[FIX] Statistics.unique: Fix Sparse Return Order For Negative Numbers

[pavlin-policar]

Issue

• The statistics module implementation of unique returned incorrectly ordered arrays in case of a negative number e.g.

>>> unique(csr_matrix([-1, 0, 0, 1, 2]))
array([ 0, -1,  1,  2], dtype=int64)

>>> np.unique([-1, 0, 0, 1, 2])
array([-1,  0,  1,  2])

• The statistics module implementation of nanunique did not support any parameters that unique supported.

Description of changes

Fixes above issues. This PR depends on #2698 for the dense_sparse testing utility.

Includes

• Code changes
• Tests
• Documentation

[codecov-io]

Codecov Report

Merging #2572 into master will decrease coverage by <.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2572      +/-   ##
==========================================
- Coverage   75.83%   75.83%   -0.01%     
==========================================
  Files         338      338              
  Lines       59545    59545              
==========================================
- Hits        45156    45154       -2     
- Misses      14389    14391       +2

[nikicc]

I only checked a0dceaf and it looks OK to me.

Other commits are a part of #2558, right?

[pavlin-policar]

Yes, this PR contains only 2 commits:

• a0dceaf fixes negative number positions
• 8add499 passes *args, **kwargs from nanunique to the unique function call.

(I've also squashed the last commit - pylint fixes into the last one for a cleaner history. No actual code changes were made)

[nikicc]

OK, then when #2558 is merged this just needs a rebase and IMO it's ready to merge.

[nikicc]

Please rebase this as #2698 has been merged.

[pavlin-policar]

@nikicc Rebased.